We have arbitrary binomials $X_n$ with $EX_n\to \infty$ and $\frac{Var(X_n)}{EX_n^2}\to 0$ . Show that



1) Let $X_n=\sum_{k=1}^n Y_{n,k}$, where iid $Y_{n,k}\sim Bern(p_n)$ then

$Ee^{i\frac{t}{\sqrt{n}}\frac{Y_{n,k}-p_n}{\sqrt{p_n q_n}}}=1+0-\frac{t^2}{2n}\cdot 1+o(\frac{t^2}{n^2}),$

since $E[(\frac{Y_{n,k}-p_n}{\sqrt{p_n q_n}})^2]=1$. So by independence

$$Ee^{it \frac{X_n-EX_n}{\sqrt{Var(X_n)}}}=(1-\frac{t^2}{2n}\cdot 1+o(\frac{t^2}{n^2}))^n\to e^{-t^2/2},$$

since $o\left(\frac{t^2}{n^2}\right)n\to 0$.

But I didn't use any of the two limits i.e.

$np_n\to \infty$ and $\frac{1}{1+\frac{np_n}{q_n}}\to 0\Leftrightarrow \frac{np_n}{q_n}\to \infty$. So there is probably an error.

2) The $X_n=\sum_{k=1}^n Y_{n,k}$, where iid $Y_{n,k}\sim Bern(p_n)$ is set up for Lindeberg-Feller theorem since for each n the $Y_{n,k}$ are independent. We have $$\sum_{k=1}^n E \left(\frac{Y_{n,k}-p_n}{\sqrt{np_n q_n}}\right)^2=1$$ and

$$\sum_{k=1}^n E(\frac{Y_{n,k}-p_n}{\sqrt{np_n q_n}})^21_{|\frac{Y_{n,k}-p_n}{\sqrt{np_n q_n}}|\geq \varepsilon}$$

$$=\frac{1}{np_n q_n} \left[np_n^2 P\left(1\geq \sqrt{\frac{nq_n}{p_n}}\varepsilon,Y_n=0\right)+nq_n^2 P\left(1\geq \sqrt{\frac{np_n}{q_n}}\varepsilon,Y_n=1\right)\right]$$

$$=p_nP(1\geq \sqrt{\frac{nq_n}{p_n}}\varepsilon)+q_n P(1\geq \sqrt{\frac{np_n}{q_n}}\varepsilon).$$

Here the problematic term is $p_n P(1\geq \sqrt{\frac{nq_n}{p_n}}\varepsilon)$ because if $q_{n}\to 0$ like $\frac{1}{n^2}$ then the limits are satisfied but $$p_n P(1\geq \sqrt{\frac{nq_n}{p_n}}\varepsilon)\to 1.$$

Can you provide some hints and corrections? Please not the solution. Thanks


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.